If 2024 was the year of the AI pilot and 2025 was the year of integration, 2026 is shaping up to be the year the invoice arrives.
As enterprises move from experimental sandboxes to full-scale commercial deployment, a new, uncomfortable reality is setting in for customer experience leaders: “intelligence” is expensive.
While the promise of AI-driven personalization and automated support is starting to deliver, the underlying costs of the hardware and software infrastructure required to power it are becoming harder to ignore. For tech buyers, that raises questions about the sustainability of AI applications in customer-facing settings.
As the costs of compute and inference skyrocket, the ripple effects are poised to disrupt pricing models, vendor stability and the pace of adoption.
Is Your CX Strategy Ready for the Real Cost of AI?
The appetite for AI in the enterprise is voracious. Recent industry statistics suggest that upwards of 80% of Fortune 500 companies have committed to integrating GenAI into their workflows by the end of 2026.
More than 95 percent of enterprises plan to use GenAI APIs or models, and/or deploy GenAI-enabled applications in production environments by 2028, according to Gartner research. And as enterprises move beyond chatbots and deploy agentic AI, 82 percent of executives plan to adopt agents to handle tasks within the next 1-3 years, according to a report by the World Economic Forum in collaboration with Capgemini.
In the customer experience space, leaders are banking on large language models (LLMs) to drive contact center efficiency, transform marketing efforts and deliver personalized, end-to-end customer journeys at scale.
But the growing cost of AI services could threaten the widespread adoption of those applications. As James Mackay, Regional Sales Manager at conversational AI firm Rasa, told CX Today, while enterprises have capitalized on free and low-cost subscriptions “the ultimate cost of AI” could come home to roost.
While OpenAI reported last week that its annual recurring revenue (ARR) jumped to $20BN last year from $6BN in 2024, it will need to bring in well over $100BN to break even.
The company’s plan to start testing ads, which CEO Sam Altman had previously said would be a last resort, suggests it is under pressure to diversify its revenue to cover high infrastructure costs.
Mackay pointed out the risk to CX operations of vendors potentially hiking prices to close their profitability gaps:
“The cost is still far cheaper than it actually is to deliver… Hopefully they’re not working towards getting people on the platform and then charging loads of money, because that will stop AI actually progressing.”
The Inference Iceberg
The primary culprit is inference costs. In the early days of the AI boom, the focus was on the cost of training models. That requires a massive, one-time capital expenditure to teach an LLM how to think. But for customer-facing applications, which can run around the clock interacting with millions of customers, the real cost is inference: the computational power consumed each time a model generates a response.
“Training, that’s a one-time billed cost, but where we’re moving is to inference, which is the ongoing operating cost of actually running AI in the real world… Training creates capability, inference determines profitability,” Lo Toney, Founding Managing Partner at Plexo Capital, told CNBC recently.
“Inference economics are going to be important to watch for 2026.”
Unlike traditional software, where the marginal cost of serving a user is negligible, every interaction with an LLM burns electricity and processing cycles. As customer experience use cases scale from hundreds of beta testers to millions of active customers, these inference costs can compound, creating unpredictable operating expenses that many enterprises haven’t budgeted for.
The stress on the system is clear at the top of the food chain. The major players providing the infrastructure, including Google, Amazon, Meta, Microsoft, are scrambling to fund the capacity required to keep the lights on. Wall Street consensus estimates for 2026 capex have been repeatedly revised higher, climbing to $527BN at the end of 2025.
OpenAI, the poster child of the revolution, faces a stark reality. Subscription revenues alone are reportedly struggling to cover the costs of operation and compute. The gap between user fees and the cost of processing queries is forcing a re-evaluation of how sustainable current pricing models really are.
This financial pressure is reshaping corporate strategies.
Amazon has made headlines for laying off workers in established divisions, a move widely interpreted as a strategic pivot to free up capital for its massive AI capacity investments. Meanwhile, Oracle has faced its own headwinds, navigating a loss of investor confidence related to its exposure to capacity deals with OpenAI that highlights the fragility of the supply chain.
Controlling and Optimizing Enterprise AI Spend
Enterprises are aware of the risk of AI costs spiraling. Around 84% of leaders believe AI costs are negatively affecting their gross profit margins by over six points, according to Mavvrik. For a typical SaaS business operating at 80 percent, that translates into significant financial loss.
“AI is no longer just experimental—it’s hitting gross margins, and most companies can’t even predict the impact,” said Ray Rike, CEO of Benchmarkit.
“Without financial governance, you’re not scaling AI. You’re gambling with profitability.”
AI use is racking up unexpected costs, as when teams introduce AI agents, they are struggling to track which agents are consuming resources. As said Žilvinas Girėnas, Head of Product at nexos.ai, put it:
“The paradox is that the same tools that help teams work faster also risk breaking cost control. Speed and visibility must go together, or it leads to chaos and unexpected expenses.”
Without sufficient oversight, no one knows which agents are active or who approved them; teams create overlapping workflows, leading to wasted spending; AI agents operate across systems without robust security checks; and finance teams cannot forecast or control costs because they lack insight into actual resource use.
How CX Buyers Should Prepare for Cost Volatility
As we move deeper into 2026, customer experience tech buyers need to be wary. The volatility in the infrastructure layer will likely trickle down to the application layer eventually.
Buyers should anticipate:
- Usage-based volatility as vendors potentially move away from flat-rate pricing towards consumption models to pass on inference costs.
- Service tiers may throttle intelligence so that faster, smarter models are gated behind premium enterprise subscriptions.
- Startups that wrap thin interfaces around major LLMs may not survive the cost crunch.
To keep costs from getting out of hand, Girėnas outlines four basic guardrails leaders should put in place. It starts with building a clear inventory of AI agents along with a “shared control plane” that gives teams room to build while still keeping a close eye on what agents are doing in production. Leaders should manage agents the way they manage people and tie agents directly to cost signals, so that teams can see usage patterns and catch budget issues early.
“The goal isn’t to slow down your fast-moving teams. It’s to let them build quickly while keeping track of what they’re building, who they’re giving access to, and what it costs.”
“This is the difference between a scalable platform and infrastructure that might become a liability,” Girėnas added.
The technology is ready for the enterprise. The question remains: is the enterprise ready to pay the real price for it?
